Refactor requantization methods

[Cesar199999]

@Antonio95 and I noticed that several functions in the quantization module share a lot of code. We came up with a couple issues that could be fixed in order to reduce the amount of repeated code:

• The requantize_fc_ntafz and requantize_fc_nte functions are almost identical. The only difference between them is the rounding method used.

• The following piece of code is repeated in almost all functions:

partial_ord_clamp(x, ST::MIN.into(), ST::MAX.into())
    .try_into()
    .map_err(|_| "Unable to convert Large Type to Small Type")
    .unwrap()

Once dimitris/requantize-bmm-ref is merged, we can start working on these issues.

[Antonio95]

Thanks! Since probably only Dimitris and I know about it, here's a piece of terminology we have coined: the "core" of the requantisation is the step that takes the i32 input and scales it in some way. After that, the zero point is added and the result is clamped. As it turns out, all requantisation methods share this second step and so only the code of the core is different, which is what you are observing :) There might be a better design reflecting this fact of life